1. Introduction

Chronic pain is a prevalent, debilitating, and complex disorder with widespread health implications.19 Patients with chronic pain frequently report a decline in cognition and memory, functions attributed to the hippocampus.14,27,66,90 There is mounting evidence that chronic neuropathic pain adversely impacts the hippocampus and its histologically distinct subfields Cornu Ammonis (CA1-CA4), subiculum, and dentate gyrus (DG) and functionally distinct subregions head, body, and tail.8,39,50,51,88,92

Structural magnetic resonance imaging (MRI) provides macroscopic assessments of hippocampal structures. Hippocampal subfield volumes are reduced in different chronic pain conditions.3–5,73,92 Similarly, hippocampal neurogenesis is blunted in chronic pain,5,28,41 which may contribute to abnormal hippocampal structure and function seen in chronic pain.28,71,80 However, there is a paucity in our understanding of the relationship between abnormal molecular environment and macrostructural change.

Diffusion tensor imaging (DTI) is an MRI technique that utilizes localized water molecule diffusion to assess tissue microstructure.35 Although DTI has been frequently utilized to assess white matter structures, an emerging application of DTI involves assessment of gray matter microstructure.61,77,97 Diffusion tensor imaging–derived diffusivity metrics have important physiological correlates. For instance, mean diffusivity (MD) quantifies overall water molecule diffusion and is related to local water content, neuroinflammation, and tissue cytoarchitecture.10,17,24,58,59 Comparatively, fractional anisotropy (FA) quantifies direction-dependent diffusion and may be reduced in the hippocampus due to multiple fibers with differing directionality.17,24,59,94 Abnormal hippocampal diffusion has been studied in a variety of conditions, including epilepsy,52,101 multiple sclerosis,16 and Alzheimer disease.31,65 In addition, hippocampal diffusivity metrics are sensitive to physiologic aging and may provide a more sensitive assessment of mild cognitive impairment (MCI) than volumetric analyses.2,24,70,76,87 These studies highlight the capability of DTI to detect microstructural hippocampal changes that precede significant volume loss. However, no prior study has reported the impact of chronic pain on hippocampal microstructure.

This study utilizes trigeminal neuralgia (TN) to investigate hippocampal microstructure in chronic pain. Trigeminal neuralgia is one of the most prevalent chronic orofacial neuropathic pain conditions.42,54,102 Uniquely, TN pain is characterized by stereotypical features, unilateral presentation, and a lack of negative symptoms including sensory loss. We recently reported that patients with TN have a subfield specific reduction in hippocampal volume,92 mostly driven by female patients.92 In this study, we utilize TN as a model to assess in vivo hippocampal diffusion to delineate chronic pain-mediated hippocampal microstructural abnormalities.

Specifically, we aim to (1) investigate diffusivity metrics in the hippocampus and its functional and structural subfields in patients with TN, (2) examine the impact of sex on diffusivity metrics in the hippocampi of patients with TN, (3) evaluate the impact of pain duration and severity on patients with TN diffusivity metrics, and (4) assess the association between the age of patients with TN and hippocampal diffusivity metrics. We hypothesize that FA will be reduced in hippocampal subfields of patients with TN, particularly the subfields responsible for neurogenesis. Similarly, we hypothesize that there will be sex-specific abnormalities in hippocampal diffusion consistent with our previous work on TN hippocampal volume. Prolonged pain duration and severity is hypothesized to be associated with abnormal diffusion in patients with TN. Finally, the age of patients with TN is hypothesized to be associated with reduced FA.

2. Methods 2.1. Ethics

This retrospective study was approved by the University Health Network (UHN) Research Ethics Board. No active participation was required for this retrospective study. Therefore, individual consent to be incorporated within this study was not required. The recruitment of healthy controls and image acquisition protocol was approved by the UHN Research Ethics Board. Individual informed consent was obtained from healthy participants. All MRI scans were completely anonymized before any image and statistical analysis.

2.2. Research subjects

Patients with right-sided TN (R-TN) seen at the Toronto Western Hospital were included in this study, each meeting the following criteria: (1) diagnosis of classical TN according to International Classification of Headache Disorders, third edition criteria,74 and (2) no surgical interventions before MRI. Patients with neurodegenerative disorders, TN secondary to multiple sclerosis, stroke, other chronic pain disorders, cranial tumors, other neurologic diseases, and history of known psychiatric disorders were excluded from this study. Healthy controls were recruited at Toronto Western Hospital.

2.3. Image acquisition

For all subjects, presurgical high-resolution, T1-weighted fast spoiled gradient-echo (FSPGR) anatomical and diffusion-weighted spin-echo echo planar imaging whole-head MRIs were acquired in the axial plane on a 3 Tesla GE Signa HDx scanner with an 8-channel head coil. The FSPGR MRI acquisition parameters were acquisition voxel size = 0.94 mm × 0.94 mm × 1 mm, 256 × 256 matrix (controls) and 234 × 234 matrix (patients), echo time (TE) = 5.2 milliseconds, repetition time (TR) = 12.2 milliseconds, flip angle = 20°, and field of view (FOV) = 240 × 240 mm2 (controls) and 220 × 220 mm2 (patients). The diffusion-weighted MRI acquisition parameters were 60 diffusion-encoding directions with b = 1000 s/mm2, 1 volume of b = 0 s/mm2 (b0) 1 excitation, ASSET, acquisition voxel size = 0.94 mm × 0.94 mm × 3 mm, 128 × 128 matrix, TR/TE = 17,000/88.8 milliseconds (controls) and 12,000/88.2 milliseconds (patients), flip angle = 90°, and FOV = 240 × 240 mm2. Voxel sixes are designed to have high levels of discrimination of the lateral/medial borders in the axial plane.

2.4. Magnetic resonance imaging processing

Diffusion-weighted MRIs underwent eddy current and motion artifact correction using affine transformation with individual subjects' gradient images to b0 image utilizing FMRIB Software Library (FSL) v6.0 program.99 Volumetric segmentation of cortical and subcortical structures including the hippocampus was performed by FreeSurfer v7.1.0.32 In addition, Hippocampal Subfield Segmentation protocol was utilized to delineate both structural and functional hippocampal subfields and extract hippocampal volume.44 Hippocampal segmentations along its longitudinal axis into head, body, and tail are used to address functional segmentation of the hippocampus. The b0 images were upsampled to T1 resolution, through FSL and subsequently registered to T1 anatomical images using FreeSurfer's bbregister function. Whole-brain FA and MD maps were obtained using FSL DTIFIT and upsampled to T1 anatomic resolution.99 Hippocampal segmentations were transformed into diffusion-weighted imaging space, and hippocampal subfields FA and MD values were obtained. The ipsilateral and contralateral sides were determined based on the side of pain experienced by the patients. As only patients with right-sided TN were included in the study, the right side for both R-TN patients and healthy controls was considered ipsilateral, while the left side was considered contralateral. Figure 1 shows a schematic diagram of image processing pipeline, and Figure 2 shows hippocampal subfield segmentation in diffusion MRIs.

Figure 1.:

Image processing pipeline for 31 patients with TN and 21 healthy controls with T1 anatomical images and eddy current and motion corrected DWI scans. T1-weighted and diffusion-weighted images were coregistered. T1-weighted images underwent hippocampal segmentation via FreeSurfer v7.1.0. Hippocampal segmentations were transformed into DWI space. FA and MD were extracted for hippocampal subfields CA 1 to 4 and longitudinal axis segmentations head, body, and tail. CA, cornu ammonis; DWI, diffusion-weighted imaging; FA, fractional anisotropy; MD, mean diffusivity

Automated hippocampal subfield segmentation and diffusion directionality in the hippocampus. Panel A shows axial views, with the left side images depicting T1 Structural MR and right images illustrating the closest axial section of the corresponding diffusion-weighted MRIs. Panel B shows zoom-in axial views of DWI images. The hippocampus is outlined on the diffusion images and is color-coded based on diffusion directionality. Panels C and D show a zoom-in axial and sagittal view of the bilateral hippocampi in T1, respectively. The hippocampus and its subfields are color-coded on T1 images according to FreeSurfer 7.0 segmentations. Notably, our investigation focuses on the histological subregions including CA 1, CA2/3, and CA4/DG, as well as functional subregions, including head, body, and tail of the hippocampus. CA, cornu ammonis; DG, dentate gyrus; DWI, diffusion-weighted imaging; HP, hippocampal proper.

2.5. Intracranial and hippocampal volumes

The intracranial and bilateral hippocampi volumes were extracted from the FreeSurfer segmentations. To account for variations in head size among our participants, we implemented the residual approach, as explained by Buckner et al.15 As previously described,15,73,83,94 we adjusted whole hippocampal volume using the residual method with the following formula:VOIadj=VOI – b(ICV – ICVmean)Where VOIadj is the adjusted volume of interest, VOI is the output volume from the FreeSurfer pipeline, b is the slope of the linear regression between VOI and on intracranial volume (ICV), and the ICVmean is the sample mean of the ICV. The t test analysis was utilized to compare the intracranial and whole hippocampal volume between healthy control and patients with TN.

2.6. Quality assurance

To validate the quality assurance within the imaging pipeline, we have incorporated several essential steps. First, to confirm that comparison of differently acquired image parameters is feasible, we compared the signal-to-noise ratio (SNR) in diffusion-weighted imaging (DWI) data for our 2 cohorts. We utilized the MRTrix imaging framework to calculate the noise and SNR for b0 and b1000 volumes of DWI, to ensure that the differences in the imaging protocols did not introduce bias to the data.

Finally, the segmentation and T1 to DWI registration results underwent manual inspection by 3 authors (S.H., A.N., and M.H.) to identify alignment and registration errors. The inspections were conducted independently, with the authors being blinded to each other's assessment.

2.7. Statistical analyses

All statistical analyses were done in Graphpad Prism v9.3.0 (Graphpad, 2020) and R 3.5.1.78 Age was compared at the group level using the Welch t test, to investigate their differences, as well as Bayesian Estimate Supersedes the t test (BEST),55 to investigate their similarities. These comparisons were performed between all TN subjects and healthy controls, female patients with TN age and female healthy controls, and male patients with TN age and male healthy controls. The Shapiro–Wilk test revealed that diffusivity metrics are not normally distributed (P-values >0.05). Therefore, the Mann–Whitney U test was utilized to assess non-normally distributed diffusivity metrics. The statistical analyses include (1) comparison of diffusivity metrics of TN subjects compared with healthy controls using the Mann–Whitney U test; (2) assessment of sex differences in diffusivity using the Mann–Whitney U test; and (3) assessment of pain duration, severity, and age using Spearman correlation; and (4) the χ2 test to determine a relationship between sex (male and female) and treatment group (TN and control). All statistical analyses underwent correction for multiple comparisons through Bonferroni correction with statistical significance set for P-value <0.05.

3. Results 3.1. Subject demographics and healthy control validation

Thirty-three patients with R-TN were included in this study (20 F, 11 M). The average age of patients with TN at the time of image acquisition was 50.9 ± 10.8 years (mean ± SD; F: 54.1 ± 9.0; M: 45.2 ± 11.9). Twenty-one healthy controls were included in this study (13 F, 8 M). The average age of healthy controls at the time of image acquisition was 45.8 ± 10.1 (F: 49.7 ± 8.8; M: 39.4 ± 9.2). Unequal sample sizes were utilized to maximize subject inclusion within the study. Age was not statistically different between all TN patients and healthy controls (P = 0.09), female TN patients and female healthy controls (P = 0.18), and male TN patients compared to male healthy controls (P = 0.25). Patient demographics are presented in Table 1.

Table 1 - Demographic summary of patients with trigeminal neuralgia. ID Age (y) Sex Distribution Pain severity (NRS) Medications Pain duration (y) TN01 38 F V3 10 Carbamazepine 6 TN02 39 F V2, V3 2 Carbamazepine, lamotrigine 6 TN03 40 F V1, V3 6 Gabapentin, baclofen, clonazepam, paracetamol–oxycodone 8 TN04 44 F V2, V3 10 Carbamazepine 1 TN05 47 F V1 9 Paracetamol–oxycodone 2 TN06 49 F V2 10 Carbamazepine, baclofen, pregabalin 8 TN07 51 F V2, V3 10 Topiramate and duloxetine 1 TN08 52 F V1 * Carbamazepine 3 TN09 54 F V2, V3 10 Carbamazepine, gabapentin 3 TN10 55 F V3 3 Carbamazepine * TN11 55 F V1, V2, V3 9 Carbamazepine 9 TN12 59 F V2 8 Carbamazepine, pregabalin 4 TN13 59 F V1, V2, V3 10 Pregabalin, carbamazepine acetaminophen–codeine 3 TN14 59 F V2, V3 5 Gabapentin 2 TN15 60 F V2 9 Carbamazepine 6 TN16 61 F V1, V2, V3 9 Carbamazepine, gabapentin 4 TN17 61 F V3 10 Carbamazepine, pregabalin 9 TN18 65 F V2 10 Carbamazepine, gabapentin 30 TN19 67 F V3 8 Pregabalin, baclofen, duloxetine 2 TN20 67 F V3 10 Gabapentin 1 TN21 23 M V2 5 Carbamazepine 3 TN22 29 M V3 4 Carbamazepine 2 TN23 38 M V2 10 Carbamazepine, pregabalin 1 TN24 40 M V1, V2 4 Carbamazepine, pregabalin 3 TN25 44 M V3 10 Carbamazepine, venlafaxine, pregabalin 1 TN26 46 M V2, V3 10 Pregabalin, gabapentin, carbamazepine 8 TN27 51 M V3 10 Pregabalin, baclofen 2 TN28 53 M V1 5 Lamotrigine, carbamazepine, hydromorphone 6 TN29 55 M V2 7 Gabapentin, carbamazepine 11 TN30 59 M V2, V3 9 Carbamazepine 2.5 TN31 59 M V1, V2 3 Carbamazepine, medical marijuana 12

Pain distribution indicates the branch of trigeminal nerve affected (V1: ophthalmic branch, V2: maxillary branch, V3: mandibular branch, *: lost in follow-up).

NRS, numerical rating scale; TN, trigeminal neuralgia.

Age comparisons did not reveal any statistically significant differences between the patients with TN and healthy controls. All corrected P-values for the Welch t test comparing age between patients with TN and health controls were above 0.05. In addition, BEST tests comparing age between patients with TN and healthy controls, female patients with TN and female healthy controls, male patients with TN and male healthy controls revealed 95% Highest Density Interval of true differences which included zero. As such, age is not only statistically different between groups but also statistically similar. The age comparison results are summarized in Table 2. In addition, the χ2 test to investigate a relationship between sex (male and female) and treatment group (TN and control) did not reveal a statistically significant difference (P = 0.85).

Table 2 - Age comparison summary between healthy controls and patients with trigeminal neuralgia. Age Mean ± SD (y) t test comparison BEST comparison Healthy controls vs patients with TN Controls: 45.8 ± 10.1
TN: 50.9 ± 10.8 P-value = 0.09 95% HDI includes zero Female healthy controls vs female patients with TN Controls: 49.7 ± 8.8
TN: 54.1 ± 9.0 P-value = 0.18 95% HDI includes zero Male healthy controls vs male patients with TN Controls: 39.4 ± 9.2
TN: 45.2 ± 11.9 P-value = 0.25 95% HDI includes zero

BEST, Bayesian Estimate Supersedes the t test; HDI, highest-density interval of true differences; TN, trigeminal neuralgia.

3.2. Imaging pipeline and quality assurance

The t test analysis comparing the noise and SNR between healthy controls and patients with TN did not show a statistically significant difference (results reported in Table 3). We further confirm that the DWI images for healthy controls and patients with TN share the same b-value, resolution, gradients, TE, and FOV and are acquired on the same type of scanner (GE Signa). Considering the absence of any SNR disparities, we are confident that the different protocols did not introduce any bias into the data.

Table 3 - Noise and signal-to-noise ratio comparison in diffusion-weighted images. Mean ± SD t test comparing healthy controls and patients with TN Noise Controls: 5.07 ± 0.44 s/mm2
TN: 4.95 ± 0.77 s/mm2 P-value: 0.52 (NSD) b0 SNR Controls: 356.39 ± 33.1
TN: 376.22 ± 63.30 P-value: 0.20 (NSD) b1000 SNR Controls: 104.79 ± 10.24
TN: 108.63 ± 16.36 P-value: 0.35 (NSD)

NSD, not statistically significant different; SNR, signal-to-noise ratio; TN, trigeminal neuralgia.

3.3. Trigeminal neuralgia is associated with specific hippocampal diffusivity abnormalities and volumes

Patients with TN had significantly reduced FA in bilateral whole hippocampi (Pcontra = 0.01, Pipsi = 0.02) (Fig. 3). Mean diffusivity values were not significantly different between patients with TN and healthy controls (Table 4).

Figure 3.:

Hippocampal FA in 21 HCs (delineated by color bars) and 31 patients with right-sided TN (delineated by white bars). Each small diamond represents 1 subject. Ipsilateral refers to the right side of the brain and contralateral refers to the left side. Bars indicate the median value, and error bars delineate the spread between the 25th (Q1) and 75th (Q3) percentiles. The Mann–Whitney U test was used for comparing patients and controls (*P < 0.05). Patients with TN have bilaterally lower FA compared with healthy controls. FA, fractional anisotropy; HC, healthy control; TN, trigeminal neuralgia.

Table 4 - Summary of corrected group comparisons between right-sided trigeminal neuralgia subjects and healthy controls. Region Side Fractional anisotropy Mean diffusivity TN (mean ± SD) Controls (mean ± SD) Corrected P values TN (mean ± SD) Controls (mean ± SD) Corrected P values Structural subregions  Hippocampus Ipsi 0.1696 ± 0.0204 0.1804 ± 0.0177 0.02* 0.001085 ± 0.000081 0.001081 ± 0.000088 0.88 Contra 0.1708 ± 0.0195 0.1808 ± 0.0135 0.01* 0.001056 ± 0.00008 0.001064 ± 0.000085 0.68  CA1 Ipsi 0.1511 ± 0.0200 0.1618 ± 0.0215 0.28 0.001099 ± 0.00011 0.001096 ± 0.00010 >0.99 Contra 0.1553 ± 0.0233 0.1640 ± 0.0148 0.076 0.001051 ± 0.000084 0.001039 ± 0.000075 >0.99  CA2/3 Ipsi 0.1890 ± 0.0377 0.2029 ± 0.0393 >0.99 0.001227 ± 0.000139 0.001212 ± 0.00017 >0.99 Contra 0.1886 ± 0.0329 0.2196 ± 0.0331 0.006** 0.001188 ± 0.000144 0.001153 ± 0.000134 >0.99  CA4 Ipsi 0.1586 ± 0.0329 0.1713 ± 0.0236 0.052 0.001005 ± 0.000083 0.000983 ± 0.000077 >0.99 Contra 0.1560 ± 0.0243 0.1787 ± 0.0196 0.0011** 0.000982 ± 0.000081 0.000962 ± 0.000072 >0.99  Subiculum Ipsi 0.1476 ± 0.0167 0.1533 ± 0.0142 0.76 0.00094 ± 0.000068 0.000949 ± 0.000075 >0.99 Contra 0.1513 ± 0.0153 0.1490 ± 0.0142 >0.99 0.000889 ± 0.000053 0.000929 ± 0.000076 0.53 Functional subregions  Head Ipsi 0.1558 ± 0.0168 0.1620 ± 0.0192 >0.99 0.001066 ± 0.000093 0.001061 ± 0.000087 >0.99 Contra 0.1566 ± 0.0181 0.1616 ± 0.0156 >0.99 0.001025 ± 0.000081 0.001008 ± 0.000069 >0.99  Body Ipsi 0.1805 ± 0.0312 0.1978 ± 0.0219 0.04* 0.001118 ± 0.000090 0.001131 ± 0.000103 >0.99 Contra 0.1811 ± 0.0240 0.2035 ± 0.0190 <0.001*** 0.001087 ± 0.000094 0.001124 ± 0.000113 >0.99  Tail Ipsi 0.1885 ± 0.0244 0.2016 ± 0.0232 0.56 0.001075 ± 0.000117 0.001053 ± 0.000116 >0.99 Contra 0.1918 ± 0.0273 0.1935 ± 0.0269 >0.99 0.001086 ± 0.000134 0.001125 ± 0.000146 >0.99

CA, cornu ammonis; TN, trigeminal neuralgia.

*P < 0.05, **P < 0.01, ***P < 0.001.

Analyses of hippocampal subfields revealed subfield-specific diffusivity differences between R-TN and healthy controls. Patients with TN had significantly reduced FA in contralateral hippocampal subfields CA2/3 (Pcontra = 0.006) and CA4 (Pcontra = 0.001). With respect to functional hippocampal subdivisions, our evaluation depicted a bilateral decrease in TN hippocampal body FA (Pcontra < 0.001, Pipsi = 0.04) (Fig. 4). Subfield MD values were not statistically significant between patients with TN and healthy controls (all P-values >0.05).

Figure 4.:

Hippocampal subfields FA in 21 HCs (delineated by color bars) and 31 patients with right-sided TN (delineated by white bars). Each small diamond represents 1 subject. Ipsilateral refers to the right side of the brain and contralateral refers to the left side. Bars indicate the median value, and error bars delineate the spread between the 25th (Q1) and 75th (Q3) percentiles. The Mann–Whitney U test was used for comparing patients and controls. All reported P-values are corrected for multiple comparison using Bonferroni correction (n.s. not statistically significant, *P < 0.05, **P < 0.01). CA, cornu ammonis; FA, fractional anisotropy; HC, healthy control; TN, trigeminal neuralgia.

The corrected hippocampal volumes based on the intracranial volume were bilaterally smaller in patients with TN compared with healthy controls (P-values: ipsilateral <0.001, contralateral <0.001). These results are in line with and replicate the previously reported finding of smaller hippocampal volumes in patients with TN compared with healthy controls.73 In addition, the intracranial volume was not statistically different between patients with TN and healthy controls (P-value = 0.08).

3.4. Hippocampal diffusivity abnormalities are driven by female patients with trigeminal neuralgia

Abnormal diffusion was restricted to female patients with R-TN. Female patients with TN displayed a significantly reduced FA in the whole ipsilateral and contralateral hippocampus compared with healthy controls (Pcontra = 0.01, P